ML engineers who need a standard reproducible way to containerize and ship models.

Cog is an open source tool from Replicate that packages a machine learning model and its dependencies into a standard Docker container with an auto generated API.

Runs locally via CLI to build images, which can then be deployed to Replicate's cloud or any Docker host.

Released by Replicate in 2022 and still actively maintained.

Removes boilerplate around Dockerfiles, CUDA versions, and API servers so model authors can focus on the model itself.

Uses Docker under the hood and generates Python based prediction interfaces.

Open source under Apache 2.0 on GitHub, no account required to build images locally, though deploying to Replicate's hosted platform needs an account.